Nutrient Document Engine MCP Server

import { MultiServerMCPClient } from '@langchain/mcp-adapters'; import { ChatOpenAI } from '@langchain/openai'; import { isAIMessage, isAIMessageChunk } from '@langchain/core/messages'; /** * Creates and initializes an MCP client for Document Engine * @param {Object} options - Configuration options * @param {string} options.serverUrl - MCP server URL (default: http://localhost:5100/mcp) * @returns {Promise<{tools: Array}>} - MCP client and available tools */ export async function createMCPClient(options = {}) { const { serverUrl = process.env.MCP_SERVER_URL || 'http://localhost:5100/mcp' } = options; const mcpClient = new MultiServerMCPClient({ mcpServers: { documentEngine: { transport: 'http', url: serverUrl, }, }, }); const tools = await mcpClient.getTools(); return { tools }; } /** * Creates a ChatOpenAI model instance with common configuration * @param {Object} options - Model configuration options * @param {string} options.model - Model name (default: gpt-4-1106-preview) * @param {number} options.temperature - Temperature setting (default: 0.1) * @returns {ChatOpenAI} - Configured ChatOpenAI instance */ export function createLLM(options = {}) { return new ChatOpenAI(options); } /** * Parses document IDs from agent response content * @param {Object} response - Agent response object * @returns {Array<string>} - Array of parsed document IDs */ export function parseDocumentIds(response) { if (!response?.content) return []; const content = typeof response.content === 'string' ? response.content : response.content.toString(); // Look for patterns like: // **FINAL_DOCUMENT_IDS: ["id1", "id2"]** // FINAL_DOCUMENT_IDS: [id1, id2] // **FINAL_DOCUMENT_IDS: [id1, id2]** const patterns = [ /\*\*FINAL_DOCUMENT_IDS:\s*\[([^\]]+)\]\*\*/, /FINAL_DOCUMENT_IDS:\s*\[([^\]]+)\]/, /FINAL_DOCUMENTS_IDS:\s*\[([^\]]+)\]/ // Handle typo variant ]; for (const pattern of patterns) { const match = content.match(pattern); if (match) { return match[1] .split(',') .map(id => id.trim().replace(/['"]/g, '')) // Remove quotes .filter(id => id.length > 0); } } return []; } /** * Handles streaming responses from agents with consistent logging * @param {AsyncIterable} stream - The agent stream to process * @returns {Promise<Object>} - The final response message */ export async function handleAgentStream(stream) { let finalResponse = null; for await (const { messages } of stream) { const lastMessage = messages[messages?.length - 1]; if ( (isAIMessage(lastMessage) || isAIMessageChunk(lastMessage)) && lastMessage?.content && lastMessage?.content.length > 0 ) { console.log(lastMessage.content); console.log('-----'); } finalResponse = lastMessage; } return finalResponse; } /** * Generic command line processor for examples * @param {Function} processFunction - The main processing function to run * @param {string} usageMessage - Usage message to display when no args provided * @param {number} minArgs - Minimum number of arguments required (default: 1) */ export async function processCommandLine(processFunction, usageMessage, minArgs = 1) { const args = process.argv.slice(2); if (args.length < minArgs) { console.log(usageMessage); process.exit(1); } try { await processFunction(args); process.exit(0); } catch (error) { console.error('Error during processing:', error); process.exit(1); } } /** * Standard document upload workflow with error handling * @param {Array<string>} filePaths - Array of file paths to upload * @param {Function} uploadFunction - The upload function from fileHandlers * @param {string} processName - Name of the process for logging * @returns {Promise<Array<string>>} - Array of uploaded document IDs */ export async function uploadDocumentWorkflow(filePaths, uploadFunction, processName = 'process') { const documentIds = await uploadFunction(filePaths); if (documentIds.length === 0) { throw new Error(`No documents were successfully uploaded. ${processName} cannot proceed.`); } console.log(`✅ Uploaded ${documentIds.length} documents. Starting ${processName}...`); return documentIds; }